Performance evaluation of artificial neural networks developed for estimation of fatigue behavior of low-alloy steels

In the literature, numerous empirical and machine learning-based techniques for estimation of strain-life fatigue parameters from monotonic properties have been proposed. Existing machine learning-based methods are evaluated in different manners, making their comparison difficult. Most authors use metrics such as root mean square error RMSE and neglect fatigue life estimations criteria, or use only conventional error criterion E_f(s). In this study, ANNs developed for estimation of fatigue parameters have been evaluated regarding their accuracy and applicability for estimation of low- and high-cycle fatigue lives of low-alloy steels, further divided into low- and high-strength subgroups.